Readings:
The article "Secure Spread Spectrum Watermarking for Multimedia" is quite interesting. The method of choice discussed in the article is to use a Spread Spectrum or a type of Spread Spectrum, that creates a watermark in media such as audio, image and video. A watermark is like a hidden copyright emblem that transmits a signal that contains relevant information about the media that it is in or transmitting from. The Watermark has to meet a number of criteria to withstand tampering of it.
It is interesting how involved watermarking is, to maintain its functionality. The article is quite informative, however at times it becomes somewhat repetitious and at times unclear. It appears to me that the introduction along with a few of the details within the paper would have been sufficient for the article and at the same time get the point across.
In the first few pages, I thought that implementing a randomized location for the watermark would eliminate the problem of it easily being determined. This thought, however, was changed since the watermark in legal cases might have to be retrieved. The issue of the potential problems that occur when trying to secure the watermark was reiterated so many times that I started picking out situations and maybe the reasons why some methods wouldn't be very successful. Some explanations of methods were like a roller coaster where at first the method appeared robust and then, just minute processes of taken by attackers would prove me wrong.
This topic is interesting to me because it brings to my mind the idea of subliminal messages being placed in media to create certain effects in humans. Since we have perceptual capacities in our audio and visual senses, is watermarking using the spread spectrum or a method close to this used in visual and audio media? I also like the part about destroying the data if an attack occurs because in the future, it can ensure the autheneticity of programs that I wite.
Digital Watermark for Multimedia documents(image,sound,video) is a new signal processing technique especially useful in protecting copyright ownership. It is an identification code that is permanently embedded in the data even after the data undergo any decryption process.
The paper goes over many previous work related to this area and finds disadvantages for all of them, such as easy to be circumvented, susceptible to attack by filtering and redigitization, etc. Then the paper discusses several important factors in making the digital watermark,i.e. the watermark should be unobtrusive, be robust to common signal processing, common geometric distortions and subterfuge attacks, be universal and be unambiguous. These factors are pretty comprehensive for watermark making. So How to achieve it? The main idea of the paper is to place the watermark in perceptually significant regions of the data which make the watermark hard to remove. The reason is that modification of perceptually significant components of a signal results in perceptual distortions much earlier than if the modifications are applied to perceptually insignificant regions. The paper brings up a technique called spread spectrum coding which is based on spread spectrum communications. The trick is to spread a narrowband signal over a much larger bandwidth such that the signal energy present in any single frequency is imperceptible. This technique ensures a large measure of security against unintentional or intentional attack.
The structure and procdure of the watermark is introduced by the paper, several experimental results are also showed in the paper. However, it seems more mathematic methods, further research and experiments are needed to make it better. For example, to identify perceptually significant components based on an analysis of the image and the human perceptual system and might also include additional considerations regarding the relative predictability of a frequency based on its neighbors, How to deal with larger system if it contains a textual image, etc.
Cox, Kilian, Leighton, and Shamoon present a new way to digitally "watermark" digital images that can also be applied to other media such as sound and video. This is especially useful today as companies are trying to spread onto the Internet without leaving copyrighted material open to unlimited electronic duplication. While this is a relatively new field, they seem to have broad knowledge of earlier works which they summarize and critique briefly.
Earlier work to digitally watermark images was based on hiding data in the less noticeable high frequency components of the data. As the authors point out, this is extremely vulnerable as most transforms such as blurring or shrinking remove or change this data. As an alternative, they suggest hiding the data within the low frequency bands which are less vulnerable to such changes without excessive noise. Since the low frequency bands are more visible, which makes them more resistant to tampering, they borrow from spread spectrum communications and divide the signal among all of the low frequency bins. Since the signal is spread out, individual parts are much smaller and less noticeable. They are also harder to destroy since they cover all of the more noticeable frequency band.
The paper seems technically well based and all the math I could follow was correct, if obscure. Their threshold for significance seems very specific for each watermark, but they do comment that it may increase for higher numbers of watermarks on file. While this will account for increased watermark density, it does not seem to address my original question but it is a minor detail though. They are also vague about how the distribution they use is more resistant to collusion but they do have a working example so it apparently works. More explanation would be appreciated.
Overall, this paper was well written and easy to follow and looks like it will lead to much new work in digital watermarking.
The authors present a new algorithm for the insertion and detection of a 'watermark' to protect copyrighted digital images, audio, and video data. The algorithm presented appears to be more robust than other currently proposed methods of encoding the watermark in digitized media. The authors further make the case for an algorithm that is applicable to various domains of digital data from images to audio to video.
This paper does a very good job of presenting most of the issues that need to be addressed by developers and users of any digital 'watermarking' algorithm. It is true that the issue of copyright protection for digitized media is becoming an important one in our society. The algorithm appears to deliver on its promises to be robust under a number of different processing method and to different types of noise. The drawback of the algorithm appears to be that no outsider can authenticate the watermark without the original watermark and this presents a significant problem. Part of the importance of copyright enforcement is that buyers can check that what they are buying is indeed authentic by looking at a front page, in the case of a book. It might also be helpful to prove the existence of a watermark alone as a partial level of authentication. Currency often includes watermarks and the lack of one is itself a proof of illegality. This algorithm, though, does seem to resolve many issues of embedding watermarks in images. One would wish for slightly more data on the effect of filtering of the image and for some more graphic representations so that perceptual acceptability could be judged. A simpler image( a box) would be helpful. I was left wondering how the alpha values would computed and how computationally expensive would this processing be since it appears that this processing has to be done with each image.
Several small details mar the otherwise very concise presentation of this paper. Page seven, " prior art protocols" seems to be a typo. And references to Lenna, though her image does not appear, are annoying. Page thirty-one, "may prefereble" is another obvious typo. The authors do not present any results or speculations on the use of their algorithm on non-static image data despite presenting their algorithm as being applicable to other media.
The author gives a number of criteria that a watermark should have in order to be effective. "It should be unobtrusive, robust, immune to common signal processing, immune to geometric operations, immune to subterfuge attacks, universal and unambiguous." The paper claims that the major problem with digital watermarking in the passed has been that the technique that were imploded were not robust enough to survive even nonmalicous modifications to the image. That is the watermarks used were not immune to operations such as rotation, translation , cropping and scaling. Such watermarks were placed in the high frequency areas of the image and were vulnerable to lowpass filtering.
The technique discussed in the paper needs to be explained in more detail. It at bests sketches the procedure that one would use. It is based on a sequence or real numbers. As far as I can tell this sequence is derived from the normal distribution of the pixels in the image. The basic idea is that you take out a sequence of numbers to which you insert watermark sequence then, you put this modified sequence back into the image.
the article describes a (then) new digital watermakring method which can be applied to audio, image, and video data. the new idea which gives rise to this new method is that the watermark should be placed in the most perceptually significant area of the image so that any attempts at changing/removing the watermark will also result in severe distortion of the original data. this is done by placing the watermark in the most significant spectral components of the data. the article shows that this technique is robust to a number of different image-distorting operations including d/a or a/d conversion, resampling, requantization, dithering, rotation, scaling, and cropping.
the article asserts that, in order for a digital watermarking method to be effective, it must be perceptually invisible, difficult to change or remove using any of the image processing functions listed above as well as others. and it should also be applicable to all media types under consideration. with these criteria in mind, the article goes on to discuss a number of previously devised methods for digital watermarking, all of which fall short of one or more of the criteria listed above.
The procedure for producing the watermark is described in general... for a message consisting of a sequence of n real numbers, the numbers are placed in the n frequency components of the image with the highest magnitudes (meaning they are the most perceptually significant). the spectrum analysis is computed using the discrete cosine transform, although the article states that other methods such as the FFT might work just as well. for the experiments described in this paper, the n integers are chosen so that they form a normal distribution, with mean = 0. a variety of possibilities are examined for the equation which is to be used to mathematically "place" the information in the message into the spectral components. the equation should be invertible, and produce results roughly proportional to the magnitude of the frequency coefficient. the mathematical specifics of the procedure are basically described, but not in any great detail.
the remainder of the article reports the results of a number of experiments where watermarked images are altered, and then the watermark is extracted and compared with the original data. as the author said, this technique is robust to nearly all of the image manipulation techniques mentioned, as shown by the experimental results. according to the criteria proposed for what makes an effective watermarking scheme, this new procedure blows away all of the previous works mentioned in terms of all around robustness, and satisfaction of the criteria.
This paper describes a very robust digital watermarking technique which may be applied to both audio and visual data. The authors clearly define their goal, as a list of requirements for a "good" watermarking method. They then provide a discussion of existing techniques, and point out weaknesses in each. Since placement of the watermark solely in the high-frequency band of the image is prone to image processing operations, as well as intentional attacks, the idea is to spread the watermark out over many frequencies. Therefore, each frequency "bin" contains only a small portion of the signal, and hopefully will not affect the image visibly. A perceptual mask is used to ensure this. Experiments show that the watermarks stand up to scaling, JPEG compression, dithering, cropping, and some malicious attacks.
I was convinced by this paper that the watermark is practically immune to accidental removal by image processing techniques. It is not clear to me whether or not a real "attack" would be possible. At any rate, I was amazed at how much more robust it seems than most other watermarking methods. Though I see the motivation for advancement in this area, I would not like to see every piece of digital media bearing these "barcodes"; an uncle of mine who works for an art history department is facing a swamp of copyright infringement problems putting together a multimedia database.
Secure Spread Spectrum Digital Watermarking was an interesting paper in the way that it described all the previous attempts at water marking, and why their strengths and weaknesses.. The authors propose a method of placing watermark data within the visually most significant frequencies of the image. The reason for doing this is that by tampering with the most significant frequencies of an image you may destroy the watermark, but you will probably also drastically reduce the image fidelity in the process..
Even though this paper is extremely easy to understand becuase of the straightforward writing of the author, the paper lacks any detail whatsoever on implementation. First of all there are no clear descriptions on how to implement a perceptual mask, nor is there any information on how to encode the data of the watermark, nor is there any information on how their watermark detector works.
At the conceptual level, this is good first reading on digital watermarks, but in implementation it leaves a lot to the imagination of the reader..
This paper describes a novel approach to digital watermarking (i.e. embedding hidden literal information into multimedia data) and compares it with other approaches. In comparison with other approaches, the authors considered the common transformations (unintentional, or intentional) that could potentially invalidate a digital watermark, and concluded that a watermark has to be unobtrusive, robust, universal and unambiguous, and that only through frequency domain water marking with the perceptually significant regions, and with continuous watermark, then this requirement could be met. I like the high level comparison of watermarking algorithms presented in this paper, but the actual algorithm the authors used is not clearly presented, especially on deciding the range of "perceptually significant regions". The augument about the authors' method being practically resilient to all possible attacks is not sound. Furthermore, the images should be presented in both the spatial and frequency domain, while none is presented, the spatial images doesn't make much sense except to prove that the watermarking doesn't alter the image much (this is definitely not the key issue about digital watermarking, I think resilience to attack is much more important).
Overall, it's very enjoyable to read this paper, and I believe the direction given by this article is the correct path towards digital watermarking.
Authors I.J.Cox, J.Kilian, T.Leighton and T.Shamoon describe a digital watermarking method for use in audio, image, video and multimedia data. The basic concept of watermarking can be thought of in the following manner: I was recently in a department store purchasing clothing. I noticed a large 'button' like piece of plastic attached to each article of clothing. It differed from the smaller plastic 'tags' that were usually present. Upon inquiring about the change, the saleswoman informed me that the people had devised ways to successfully remove the smaller tags without doing harm to the clothing. This implied that the clothing could be taken from the store without being paid for and subsequently worn or resold. However, if the new larger buttons were removed by someone other than an authorized person, a large ink mark would be left in its place, thus effectively ruining the article of clothing. Of course, for digital data, the watermark should not be perceptible to the human eye or identifiable (through data manipulation) by unauthorized persons. The parallel between the two can be found in the facts that (1) a new method needed to be devised because previous methods had ceased to be effective and (2) any attempt to remove the 'security tag' (watermark for digital data or button for clothing) would effectively result in severe degradation of the original item following the attack.
Two key issues, in regards to the effectiveness of the author's proposed technique for watermarking digital data, emerged as the central theme throughout the paper. They were (1) the scheme must be frequency based to avoid degradation of the watermark due to spatial manipulation of the data (i.e. cropping) and (2) the scheme must place the watermark in significant frequency components of the data to avoid degradation of the watermark due to low-pass filtering of the data (i.e. compression schemes).
The methods employed for formulating the watermark and evaluating it's robustness appeared rather straight forward, although I found myself somewhat at a loss in their analysis of attacks on multiple versions of a watermarked data. It is my opinion that the author's sufficiently answered lingering questions concerning parameters and overall effectiveness. Nowhere in the article did they imply that their proposed technique would be 100% foolproof. Instead, they showed that the actual watermark implemented is dependent on a variety of factors associated with the original data. However, the amount by which the watermark alters the data (parameter or scaling factor a) appears to be somewhat of a daunting task to compute, subject to trial and error especially if one has multiple scaling parameters. It would be interesting to test the robustness of the method with a wider spectrum of data (i.e. video, audio.). However, the authors explicitly stated that more experimental work needs to be performed to validate the results.
The proposed method applies the watermark onto the most perceptually
significant components of the signal, using frequency-based method.
Spread spectrum coding procedure is based on the frequency domain
transformation and in the frequency domain, each coefficient has a perceptual
capacity. It is not clear, following the article, how to determine the perceptual
capacity of each frequency.
To insert a watermark X, a sequence of values are extracted from each document
and combine X to them to get an adjusted sequence, and then they are inserted
back into the document to obtain a watermarked document.
The procedures of inserting and extracting the watermark and of evaluating the
similarity of them are introduced. In some trivial point of view, there are
many typos and unclarities in this article.
In the general concern of cryptography, to find an one-way function is hard problem, because inverting the function should almost be impossible. Especially when the domain of the function is variously complicated, it is likely for an one-way function to fail in certain value of domain. Since there are many categories of attack such as cropping, rescaling, and so on, which should be considered for a watermark to be really robust and flexible, I suggest such a combinatorial way that first we find a set of trusted watermark for each category by applying the concept of the one-way function, onto the signal and then evaluate the combinations of plausible watermark from each set, so that they can be used in the combined mode for a certain type of multimedia, say image, audio, and so on. and moreover can achieve flexible and comprehensive encryption which doesn't harm to any of the categories to which attacks can be protected from.
This paper introduces the reader to a very interesting concept of protecting the electronic data by using the watermarks. What the author suggests is to insert the watermark into the document. The watermark would contain the information about the author or producer who designed and implemented this code and probably about the person who purchased it. In order to use the watermark as a legal proof it should satisfy a number of specifications. The watermark must be placed in perceptually significant regions of the data despite the risk of potential fidelity distortions. What it means that any attempt to remove the watermark would significantly effect the data and also the watermark should be robust to any attempts to alter it. This paper gives a mathematical foundation for building the watermark. Some possible algorithms for creating the watermark are given along with the experimental results which proof that the watermark is robust.
Unlike intellectral rights protection for sofeware packages or videos, the objective of multimedia watermarking is to uniquely label images or audio materials so that any unauthourized use can be unequivocally detected and proved in the court of law. This is equal to say, the use of image processing tool like xv to grab BU's logo can be detected if the logo on BU's web page is watermarked. Understandablly, this is a difficult task because of completely different context the relevent technologies are involved in. Ingemar J. Cox, Joe Killan, Tom Leighton, and Talal Shamoon 's article, "Secure Spread Spectrum Water Marking for Multimedia" (NEC Research Institute, Technical Report 95-10), offers a good discussion on the issues involved.
Essentially, the authors argue that since perceptually unsignificant component can easily be modified through signal processing, watermarking can only be done on perceptually significant part. Then there is a problem. Because the reasons why people perfer perceptually unsignificant part is that operations on perceptually significant part will affect the fidelity of the signal being protected. To overcome this problem, the authors proposed hiding a narrow band of watermark information in to a wideband of data. Because the watermark informationis spread out on the signal band, the quality of the signal is not compromised. The author especially concluded that this method is applicable to all medias including audio, image and video.
To implement their ideas, the authors made watermark structure with 1000 randomly generated numbers which obey normal distribution. Their watermark is placed in the perceptually most significant components of the image spectrum. Their watermarker withstood operations like zooming, cropping, JPEG encoding, dithering , printing, photocopying and subsequent rescanning.
The authors have successfully brought up issues related to the media watermarking technology. Their concept of watermaking the perceptually significant part is very valuable. But they stop short of giving more evidences to support their theory with other forms of media. This especially leaves readers wonder how applicable their success on operations of one image file will be. In their experiment, the authors had chosen perceptually most significant part for watermarking. This leaves readers wonder what will happen if the watermarking was made on slightly less significant part. Obviously, the degree of significance needs to be clearified if not quantified. In view of the fact that this princeple is applicable to all three forms of media ( which I agree), there should be a criteria to define the level of significance before their principle can be widely use.
The possibility of copying digital information in many easy ways and with a degree of accurateness extremely high created a complicate problem for copyright protection, this means that most of the times it is impossible to distinguish whether data is original or copied, generating evident copyrighted ownership polemic.
This work deals with the problem of watermarking general multimedia data (images, video and audio). In general a watermark is a seal placed on the image in order to identify, for example the copyright owner of multimedia data and the identity of the purchaser of one of the legally sold copy.In order to create their design, the authors considered a wide range of issues that have to do with the effectiveness of the watermarking. Some important ones include the obvious necessity of making it perceptually invisible, difficult to remove without degrading the data fidelity, robust against signal processing, geometric distortions and systematic attacks to destroy it, and unambiguous (easy to identify its source).
According to the article it can be applied to different media representations which gives a very desirable property that can make it convenient to implement. The main contribution of this work is the technique of placing the watermark in perceptually significant components of the signal. The principal reason for this is to increase its robustness against removal operations or common signal and geometric distortions. Placing the watermark in insignificant regions will decrease its possibility of survival due to the fact that most of the techniques for signal processing and manipulation can detect where this insignificant information is locate and prefer to lose it. This approach is practically based in that despite the fact that the data might potentially lose fidelity, watermarking changes must be made on perceptually significant regions of the data, specifically in significant spectral components of the signal. This will increase is robustness against distortions that irretrievable-loss-of-data algorithms could cause.
There is one basic problem against this approach, this was the reason why most of the previous techniques used high frequency marks, this is: how to insert the watermark into the most perceptual regions of an spectrum and at the same time avoid generating noticeable alterations?.This problem was solved by viewing the frequency domain of the image or sound as a "communication channel and the watermark as a signal that is transmitted through it". The watermark is spread over many frequency bins so that the energy in any bin is small and undetectable. It is ways to detect the watermark if we know its location and content by concentrating these many weak signals into a single signal with high signal/noise ratio.
A watermark could be placed in a frequency domain in a way that would make it practically impossible to perceive. For an attacker to be sure of the complete elimination of the watermark, he has to perform strong modifications, degrading the quality of the image.
Applying this method, this work computed the N X N DCT of the image and place the watermark into the n highest magnitude coefficients of the matrix (excluding the DC component). The most important point is that modifications in these locations will degrad e the image quality in a high level.
This method is one of the first that is able to deal with many of the problems that watermarking has had. Its results showed a good general performance, guarantying uniqueness of the watermark, supporting of image scaling, general coding distortion, geome tric, dithering distortion, cropping, inclusive printing and re-scanning processes. It was proved to be very robust against counterfeiting. Some weaknesses of the technique are its impossibility to work on ascii text (which is obvious), furthermore in tex tual images watermarking is easy to detect by character recognition methods.
In this article authors are presenting a new method of digital watermarking (putting a hidden signature on an image (or a sound) file, for copyrighting or other purposes. They discuss many other different techniques which were used for this purpose, most of which used least significant bits(high frequencies) to store information. These techniques are susceptible to simplest attacks which do not decrease image quality (such as lowpass filtering or cropping). The authors go on to present their technique which attempts to spread the watermark information over the most significant frequencies, so that attacks which destroy the watermark also render an image or a soundbite unusable. The main idea of the of the technique is that small changes to big frequency coefficients are impossible to see or hear.
Since low frequencies are being marked in this watermarking scheme, watermark should be resistant to such attacks are lowpass filtering, cropping, scaling, which get rid of high frequencies, but do not strongly influence low frequencies(no rigorious mathematical proof of that was presented). The scheme seems to work (according to experiment data presented), although the exact math of the comparing original and retrieved watermarks was not really developed. Also no research on the noticability of the changes has been made.
Watermarking is a very interesting topic. The paper does a good job of explaining the basic prinicples of watermarking and previous attempts. The method they describe for watermarking sounds remarkable. They state that their watermarked images are almost invulnerable to malicious attack.
They give examples of trying to distort the image and destroy the watermark, and they show that the watermark is still able to be found even after the image has been scaled, compressed into a JPEG image, dithered, and clipped down to 25% of the original image size. And even more amazing, the watermarked image was printed out, copied with a xerox machine, and then scanned back into the computer, and after all that, the watermark was still able to be found. Even watermarking the same image multiple times didn't cause interference. They watermarked the same image 5 times, and when they tested 1000 randomly generated watermarks, all 5 of the watermarks showed up clearly.
I understood most of the article very well except for section 4 where they explain how to watermark an image. I was a little confused when they were discussing the actual algorithms used. Overall the article was very interesting. Although this is all I've ever read about watermarking, it sounds like a very effective method of putting a "signature" on something and protecting software from piracy.
Stan Sclaroff
Created: Oct 4, 1996
Last Modified: Oct 5, 1996